Detection of Diabetic Retinopathy Using Deep Learning Techniques | ||||
| ERU Research Journal | ||||
| Volume 4, Issue 3, July 2025, Page 3097-3115 PDF (773.67 K) | ||||
| Document Type: Original Article | ||||
| DOI: 10.21608/erurj.2025.369875.1238 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
Amira Mofreh; Aya Adel   ; Norhan Elnady ; Mahmoud Khaled Abd-Ellah
| ||||
| Faculty of Artificial Intelligence , Egyptian Russian University, Cairo 11829,Egypt. | ||||
| Abstract | ||||
| Diabetic retinopathy (DR) is a significant problem of diabetes, leading to vision impairment and blindness if left untreated. Early detection is crucial for effective intervention. This paper uses deep learning methods to detect DR from retinal fundus images automatically. Five pretrained convolutional neural network (CNN) architectures, including VGG16, ResNet50, InceptionV3, MobileNet, and DenseNet121, were modified, retrained, and evaluated on a standard dataset. Different evaluation metrics such as accuracy, ROC, and F1-score were used to evaluate model performance. The dataset used in this project is sourced from Roboflow and is designed to detect diabetic retinopathy. The dataset is divided to training, validation, and testing with 70%, 20%, and 10% respectively. Results demonstrated that the DenseNet121 model can effectively detect DR, with the best-performing model achieving accuracy (AC), precision (PR), recall (RC), false positive rate (FPR), F1-score (F1), and ROC curve (AUC) of 93%, 91.60%, 92.25%, 6.45%, 93%, 0.98% respectively. This paper discusses these findings' implications and suggests future research directions. | ||||
| Keywords | ||||
| Diabetic retinopathy; CNN; Detection; Resnet50; DenseNet121 | ||||
|
Statistics Article View: 119 PDF Download: 166 |
||||
